OPENAI IS ADDING "DESCRIBE A VOICE, GET A VOICE" TO ITS API. NOT ANNOUNCED YET
spotted in the official openai SDK, already live on npm (7.28.0)
→ new endpoint: POST /audio/voices
→ write a text description of the voice, get back a saved voice id
→ optional script for it to read, otherwise it writes one from your prompt
→ model version "2026-10-01"
→ text designed voices only work in GPT-Live for now, not Realtime or the speech endpoint
cloning from a real audio sample is in there too, still behind a consent recording from the speaker
google shipped the same idea in gemini 3.8 flash TTS 2 weeks ago